Home → blog → What does watermarking text by LLMs mean for your brand identity?
What does watermarking text by LLMs mean for your brand identity?
by Karolina Kurcwald
So, LLMs are watermarking text they generate to comply with the EU AI Act. Good.
But there’s something particularly interesting about this. Because the LLM watermark, this time, is not some type of metadata. It’s a specific pattern of word choice, basically, a signature style for every LLM. To remove it, you have to edit the text – probably more heavily than not.
Now, think about it. A specific pattern of words and phrases used. A cadence. A tone.
Basically, it’s brand voice.
My question is: where’s yours?
What’s that LLM watermark all about?
In case you haven’t heard, Claude’s parent company Anthropic announced in August 2026 that all text generated by future models of Claude will contain a watermark to comply with the EU AI Act. This seems to have sparked a larger conversation about watermarking text by other LLMs, with, for example, Gemini using a system called SynthID-text to mark the text it creates.
The watermark isn’t some sort of a code or metadata that you can remove. In general, the LLM watermark is a pattern used to select words by the different LLMs. It’s generally undetectable by the reader – you’d need a key to decipher it.
When watermarking is used, choices are still made at random, but the source of the randomness is different. Instead of using an arbitrary random number generator to pick the next word, watermarking uses the key and a few words that come before to settle what word the model should pick. That is, the words that Claude picks are still random, but now, one can check the sequence of words and see if it’s consistent with the choices Claude would make if it was using the key. If it is, one can assign a probability that the text was generated by Claude.
In other words, unless you use what Claude generates as a very rough draft and essentially rewrite most of it, the fact that the text is generated by Claude will be detectable (if you have the key).
As Anthropic explains, when you use Claude to edit your text, the presence or absence of the watermark will depend on how much Claude inteferes with the original.
Obviously, this is helpful for legal reasons. But I have a different question (because legality doesn’t seem to stop people from throwing fully AI-generated text on their websites).
What does that mean for your brand identity?
And I think this is a much more nuanced conversation.
Now, my experience in copywriting tells me that many brands only care about their brand identity so far as the visual branding is concerned. The verbal branding, as I like to call it, is often not even an afterthought – and brands that don’t think about it, won’t mind using AI-generated text, as there’s basically no difference to any other type of text for them.
And so they continue to have brandbooks with detailed color schemes and pixel-perfect logo white space, with maybe three-four adjective as their brand voice.
You know, human, conversational, that kind of thing. Essentially, what the text I’m writing could be described with, along with most of content you can find online.
But no specific word choices. No “keys” to constructing different types of text. No words to use and not to use. Ways to address the reader, ways to talk about the product, to resolve conflict on social media.
Why would you do that? Well, the way you talk is part of your personality. And it’s the same with your brand. Would your audience know it’s you if you removed your logo and brand colors from your website? Or would it just now be Claude and the way it writes?
Are we going to have countless Claude brands now?
Because essentially, this is what it amounts to.
You can think of the watermark as a type of a voice – for Claude, Gemini, ChatGPT, etc. We don’t yet have publicly available tools to detect it (or correct me if I’m wrong), but it’s soon going to happen.
But how does your audience detect it’s you from a piece of text – on your website, your blog, your LinkedIn post?
Think about it – soon, people will know Claude wrote something for you. But they still might not be able to recognize you by the way you talk.
There will be essentially just be copies of the same kind of writing everywhere.
And even if you coach your LLM to write more like you, it will still make its choices – the ones that make up its watermark.
Think about it. (I know I already told you to think about it, but seriously, do it.)
So, what do you do about it?
Of course, there will be brands that won’t care – and I guess this post is not for them.
But if you do care, figure out your brand personality and voice first. In a way that’s actually detectable for your audience.
Then make sure anything your brand puts out there clearly show it’s you.
Then, I’ll go so far as to say it might not even matter if Claude wrote it.
I can help
Let’s talk about your brand and your voice – you can set up a quick call here to tell me all about it.
And if you’re looking for the right words for your personal brand – as a founder, expert, solo business provider, etc. – I’m soon launching “The Words for It”, a 1:1 coaching program to come up with your own unique way to talk about what you want to be known for.